OC20 bare slab energies

[R1j1t]

Hi Team,

I want to access bare/clean slabs from OC20 dataset. I had successfully completed similar exercise for OC22 however I am struggling with OC20 and have the following questions:

1. Where can I extract clean/bare slabs structures (initial and final) and their relaxed raw DFT energy. Current LMDB entries in IS2RE have e_adsorption.
2. I tried ads+catalyst to clean catalyst mapping to get IDs for the clean slab but could not find them in the database.

Current attempt

with open('oc20/oc20_data_mapping.pkl', 'rb') as f:
    '''
    {'bulk_id': 6510,
     'ads_id': 69,
     'bulk_mpid': 'mp-22179',
     'bulk_symbols': 'Si2Ti2Y2',
     'ads_symbols': '*N2',
     'miller_index': (2, 0, 1),
     'shift': 0.145,
     'top': True,
     'adsorption_site': ((4.5, 12.85, 16.13),),
     'class': 1,
     'anomaly': 0,
     'split': 'test_is'}
    '''
    data_mapping = pickle.load(f)

with open('oc20/mapping_adslab_slab.pkl', 'rb') as f:
    adslab_2_clean_slab_map = pickle.load(f)


# Sample data in train LMDB
# Data(edge_index=[2, 688], pos=[39, 3], cell=[1, 3, 3], atomic_numbers=[39], natoms=39, cell_offsets=[688, 3], force=[39, 3], distances=[688], fixed=[39], sid=1334331, tags=[39], y_init=1.4904021400000005, y_relaxed=-0.7590455899999995, pos_relaxed=[39, 3])

clean_slab_id = adslab_2_clean_slab_map['random1334331'] # random989340

print(data_mapping[clean_slab_id])
# {'bulk_id': 1605,
#  'ads_id': 33,
#  'bulk_mpid': 'mp-684799',
#  'bulk_symbols': 'K5As4',
#  'ads_symbols': '*CCH3', # I was assuming catalyst to not have any adsorbate here
#  'miller_index': (2, -1, -2),
#  'shift': 0.041,
#  'top': False,
#  'adsorption_site': ((4.25, 3.95, 24.82),),
#  'class': 1,
#  'anomaly': 0,
#  'split': 'train'}

I tried searching 989340 in the LMDB but no luck. Any help would be appreciated!

[mshuaibii]

Hi - Sorry for the confusion. Clean slabs can be found at this download link: https://fair-chem.github.io/catalysts/datasets/oc20.html#catalyst-system-trajectories-optional-download.

[R1j1t]

Thanks for the quick reply @mshuaibii. The above link looks good, however, is there a LMDB file which has the final energies? I am a bit concerned working with 151 GB of data and extracting final relaxed energies. Perfectly okay if there isn't one and I can use the above trajectories.

[mshuaibii]

Unfortunately we don't at this time so you'll need to extract them directly from the trajectories.

[R1j1t]

No worries! I appreciate your team's work and have stared this repo. Excited for future releases, cheers!